Nominal subextractions and the structure of NPs in Serbian and English

(1)

Nominal subextractions and the Structure of NPs in Serbian and

English

Monika Baˇsi´c

MPhil-avhandling Det humanistiske fakultet

Universitetet i Tromsø

Juni 2004

(2)

Acknowledgements

First and foremost, I would like to thank my supervisor, Prof Peter Svenonius for invaluable suggestions and many fruitfull discussions we had, not just while writing this thesis, but also during the two years I spent in Tromsø.

Secondly, I would like to thank my fellow students and all the teachers at the faculty for creating a great environment for studying linguistics.

I am grateful to my friends who have practically been my second family here, especially to Zhenja, Patrycja, Kaori, and Madeleine.

My warmest thanks go to Luci, who may have been far away, but was always in my thoughts.

Finally, I am deeply indebted to my parents, Katarina and Zvonimir, and my brother, Nikola for their love, support, and encouragement all these years.

(4)

1 Introduction

The present work represents an analysis of constructions referred to in the literature as Left Branch Extractions. Ross (1967) was the first to note that many languages ban movement of left branch constituents from within the noun phrase.

This is exemplified below for English and Dutch (Dutch examples are from Corver 1990).

(1) English

a. *Which did you like film?

b. *Whose did you crash car?

(2) Dutch a. *Welke

which zag saw

jij you

hond van Jan?

dog of John

‘Which dog of John’s did you see?’

b. *Jans John’s

heb have

ik I

boek book

gezien!

seen

‘John’s book I have seen.’

In order to account for the frozen character of these constituents, Ross (1967) formulated a contraint known as Left Branch Condition (LBC).

(3) Left Branch Condition¹

No NP which is the leftmost constituent of a larger NP can be reordered out of this NP by a transformational rule. (Ross 1967/86, p. 127)

However, already Ross himself recognized that the LBC is freely violated in a number of languages, especially those exhibiting relatively free word order. Par- ticularly famous in this respect are Slavic languages such as Serbo-Croatian, Pol- ish, Czech and Russian, which allow all their prenominal elements to be separated from the rest of the noun phrase.²

1Obviously, not all prenominal elements in (1) and (2) are NPs on standard assumptions. How- ever, in Ross’s original analysis, all these constituents were treated as derived from NPs. In other words, demonstratives, adjectives, and possessives are all dominated by an NP layer up to some point in the derivation.

2Serbian is the official name of the language spoken at present in Serbia and Montenegro.

Serbo-Croatian was the official name of the language before it split into Serbian, Croatian and Bosnian during the 1990s, and this is how most of the previous literature refers to it. Although I will use the name Serbian, the analysis to be presented carries over to Croatian and Bosnian as well, as all three languages behave in the same way with respect to the phenomenon under

(5)

Introduction

(4) Serbian a. Koliko

how.many su aux

otpustili fired

radnika?

workers

‘How many workers did they fire?’

b. Koju which

su aux

zatvorili closed

fabriku?

factory

‘Which factory did they close?’

(5) Czech

a. Jak siln´eho how strong

videl saw

Jan John

muˇze man

‘How strong a man did John see?’

b. Jakou which

ˇcte reads

Petr Peter

knihu?

book

‘Which book does Peter read?’

The phenomenon of LBE has received much attention in the literature. Probably the most detailed study couched in the Government and Binding framework was presented by Corver (1990). Corver assumes that parametric variation results from the interaction of the ECP and the categorial status of noun phrases. The core idea of his proposal is that Slavic languages which do not obey the LBC lack the DP functional layer. On the other hand, in languages with over determiners, D projects a minimality barrier and blocks the extraction of prenominal material.

However, I will argue that a direct extraction approach cannot be maintained despite its initial appeal. A number of important properties of split constructions are not accounted for under a Corver-style approach. Secondly, I will show that the central asumption of Corver’s analysis is not supported by empirical evidence.

The conclusion will be that the availability of LBE cannot be reduced to the presence vs absence of a DP layer.

By focusing primarily on empirical data from Serbian, I will argue that cases of apparent LBE are actually derived by extracting the non-focused material from the dominating DP/PP. Subsequently, the remnant DP/PP is fronted to the left periphery of the clause. In this respect the term Left Branch Extractions is in fact misleading, since on my assumptions the left branch element does not leave the DP. To avoid confusion, I will thus refer to this phenomenon by using the term split constructions, or XP-splits. I will show that the remnant movement analysis

discussion.

(6)

Introduction

straightforwardly captures a number of properties otherwise problematic on direct extraction approaches.

On the theoretical side, I will argue that not all movement operations can be characterized in terms of Attraction. A guiding idea of the Minimalist Program is that movement is not optional. It is a last resort operation triggered by the need to check features of lexical items in order to ensure convergence at the interface levels. Chomsky (1995) argues that what triggers movement is a morphological requirement of the target, rather than of the category which moves. The probe attracts the closest element bearing the relevant feature F.

(6) K attractsαonly if there is noβ,β closer to K thanα, such that K attracts β.

Despite its conceptual elegance, the theory of Attract raises some non-trivial questions. Attractors bearing relevant features must be identified in all cases. That this is not an easy task is evident from the fact that in some cases purely formal features need to be postulated, the role of which is only to trigger movement.

Particularly problematic for the view of displacement as a last resort strategy are what seem to be optional operations such as scrambling. The apparent optionality is a characteristic of split constructions as well, pied-piping of the entire DP being a grammatical option.

(7) a. Koliko how.many

radnika workers

su aux

otpustili?

fired

b. Koju which

fabriku factory

su aux.

zatvorili?

closed

I will argue that the movement step responsible for deriving XP-splits is not triggered by Attraction, but is brought about by the properties of the source position.

The split arises when two elements within the same phrase bear conflicting features. The assumption that this movement is of different nature will be further supported by the behaviour of split XPs with respect to island constraints.

The thesis is organized in the following way. In Chapter 2, I will give my assumptions concerning the structure of Serbian noun phrases. I will also explore arguments against the DP status of Slavic noun phrases, and conclude that there is no conclusive evidence that would require abandoning the DP-hypothesis for Slavic. Chapter 3, provides the basic descriptive facts concerning the extraction possibilities in Serbian. In Chapter 4, I review some of the previous approaches to the phenomenon under discussion. Chapter 5 forms the core of this thesis where

5

(7)

Introduction

I present my analysis of split constructions. I will first argue that split XPs are formed by two instances of movement: extraction of the non-focused material, followed by fronting of the remnant XP. I then turn to motivation underlying this type of displacement, and show that it is driven by focus considerations. I also discuss the nature of the first movement step in some detail, arguing that it is best understood as being driven by the properties of the source position. Finally, in chapter 6 I will briefly investigate the possible sources of cross-linguistic variation, and provide some compelling evidence against a Corver-style approach.

Chapter 7 summarizes the conclusions drawn from the present study.

(8)

2 The Structure of English and Serbian NPs

2.1 The Structure of English Noun Phrases

There is a lot of controversy in the literature concerning the structure of noun phrases, particularly in determining the number and type of functional projections dominating the NP. Traditionally noun phrases were analysed as maximal projections of a lexical head N⁰, as represented in (1).

(8) NP

Det N^′

N Compl

Investigation of the symmetry between clausal and nominal domains has led researchers to the conclusion that NPs are, like VPs, dominated by functional projections. Thus, Abney (1987) proposes that noun phrases are maximal projections of D, the position where articles are inserted, and assumes the following structure:

(9) DP

(Poss) D^′

D NP

On this approach, the function of the determiner is to specify the reference of the noun phrase by picking out a particular member of the noun’s extension. This parallels the function of Infl in the verbal domain, where tense locates a particular event in time. The DP-analysis of noun phrases was further used to account for certain cooccurrence restrictions. For instance, the fact that determiners cannot cooccur with pronominal possessives in English follows on the assumption that they are all heads and occupy the same syntactic position, namely D⁰. D was also argued to host the possessive morpheme ’s, which takes the possessor DP as its specifier. This directly captures the ill-formedness of the following examples:

(10) a. *this Peter’s article b. *the his article c. *that the article

One of the strongest arguments in favour of the DP-hypothesis was the behaviour of gerundive nominals in English, such as the one below (from Abney, 1987):

(9)

The Structure of English and Serbian NPs

(11) John’s building a spaceship

English gerundive nominals show properties of both nominal and verbal phrases.

They have the distribution of noun phrases, but internally they show verbal char- acteristics, for example, taking nominal complements, aspectual auxiliaries, and allowing adverbial modification. The mixed behaviour of gerundive nominals was captured by assuming the structure in (12), where the functional category D takes a VP as its complement.

(12) DP

John D^′

D

’s

VP V building

DP a spaceship

Since Abney’s influential proposal, much work was devoted to establishing the validity of the DP-hypothesis. Bringing cross-linguistic data into considera- tion, particularly strong support was provided by the presence of overt raising of N to D in some languages. Longobardi (1994), investigating the distribution of determiners and proper names in Italian, notes that adjectives can either precede or follow the proper name when the determiner is present, but when it is lacking the adjectives must follow the noun. On the other hand, in English adjectives must always precede the proper name. The contrast is illustrated below.

(13) a. (i) E is

venuto come

il the

vecchio old

Cameresi.

Cameresi

‘The old Cameresi has come.’

(ii) *E venuto vecchio Cameresi.

(iii) E venuto Cameresi vecchio.

b. (i) Old Peter has come.

(ii) *Peter old has come.

Longobardi (1994) accounts for the observed contrast by assuming that in Italian proper names may move to D by Spellout. If they fail to do so, an expletive article is inserted in D, and the proper name moves to D in LF. On the other hand, in English N raising takes place only in LF, in accordance with the Principle

(10)

Italian shows another interesting contrast. Longobardi points out that while omission of the article is not possible in (14), where the noun phrase is an argument, it is possible in (15), where the nominal expression functions as an invoca- tion (15-a) or a predicate (15-b).

(14) *(Il) the

grande great

amico friend

di of

Maria Mary

mi me

ha has

telefonato.

telephoned

‘The great friend of Mary has telephoned me.’

(15) a. Caro dear

amico, friend,

vieni come

a to

trovarmi.

visit.me b. Gianni

Gianni e is

amico friend

di of

Maria.

Mary’s

According to Longobardi, the difference resides in the fact that nominal ex- pressions in (15) are not arguments; in order for nominals to function as arguments they must have a lexically filled D. On such a view, NPs are treated as bare pred- icates, and the function of D is to convert the NP into a referrential expression, which can then be used as an argument.

The proposal for introducing a DP functional layer has received a strong empirical support by a series of studies. Thus without going any deeper into the issue, I will assume that the DP layer is present in English. Moreover, I will argue that the DP-hypothesis is also valid for Slavic NPs, which has been a matter of much controversy in the literature on Slavic. I return to this issue in the next section. I first discuss my assumptions regarding the placement of adjectives within DPs.

Attributive adjectives Much debate has centered on the status and position of attributive adjectives. On Abney’s view, adjectives are heads in the extended projection of the noun phrase. The functional head D may select AP as a complement, yielding the following structure for the example in (16).

(16) a. a cold beer

b. DP

D a

AP A cold

NP beer

9

(11)

However, this analysis proved to be unsatisfactory in many respects. One obvious problem is that adjectives within NPs can be phrasal, in other words they can be modified and they can take complements. The structure in (16) leaves no place to accomodate these elements. Consider the following Swedish example, where the adjective v¨alk¨ant ‘well-known’ is premodified by a PP (from Delsing, 1993).

(17) ett a

sedan since

i g˚ar yesterday

v¨alk¨ant well.known

faktum fact

‘a fact well-known since yesterday’

Furthermore, Svenonius (1994) notes that, if adjectives are heads, we would in- correctly predict that a degree element scopes over all the adjectives that follow it. The Norwegian example below illustrates that the degree element altfor ‘much too’ takes scope only over the adjective heit ‘hot’, there is no implication that the coffee is too strong as well (from Julien, 2002).

(18) alt-for all-too

heit hot

sterk strong

kafee coffee

‘much too hot strong coffee’

Cinque (1994) further observes that the distribution of adjectives in noun phrases closely resembles the distribution of adverbs in the VP. The speaker- oriented adjectives like probabile ‘probable’, sicuro ‘sure’, are followed by subject- oriented APs. These are in turn followed by manner or thematic APs.³ Consider the example in (19-a), and the corresponding ordering of adverbs within VP in (19-b).

(19) a. La the

probabile probable

goffa clumsy

reazione reaction

immediata immediate

alla to

tua your

lettera.

letter b. Probabilmente

probably

avranno have-3Pl

goffamente clumsily

reagito reacted

subito immediately

alla to

tua your lettera

letter

Cinque (1994) entertains two solutions to the question of where APs are generated; either they are adjoined to a maximal projection, or they are generated in distinct specifier positions of functional projections between D and NP. He con- cludes that the latter alternative is to be preferred on both conceptual and empirical reasons. First, the observed ordering of APs follows from the hierachical ordering

3Thematic (agentive) adjectives such as Italian in The Italian invasion of Albania are taken to express the external theta-role of N.

(12)

of functional projections in whose Specs APs are generated, whereas the adjunction is typically conceived as being free. Secondly, there is a limit on the number of attributive APs within DP, while adjunction positions are unlimited.⁴ Finally, if APs are generated in specifier positions, the fact that they surface to the left of the head need not be stipulated but follows from the location of specifiers.

Julien (2002) argues that Scandinavian provides evidence in favour of Cinque’s proposal. Consider the example (20), which is marginaly possible in Norwegian, but fully grammatical in some dialects of Northern Swedish.

(20) ?eit

indef-N.Sg stor-t big-N.Sg

eit

indef-N.Sg styg-t ugly-N.Sg

eit

indef-N.Sg hus house

‘a big ugly house’

The articles do not show adjectival agreement, which is spelled out by the suf- fixes on the adjectives. Julien takes this to mean that articles are not contained in APs, but are actually heads of the functional projections in which adjectives are generated.

I thus conclude that the analysis of adjectives according to which they are generated as specifiers of distinct functional projections is preferred over the analyses which treat them either as heads or as phrases adjoined to NP. Moreover, if my analysis of Serbian extraction facts is on the right track, it lends further support to Cinque’s proposal.

In conclusion, considering all the arguments presented in this section I will assume that the structure of English noun phrases is as shown in (21). Determiners and pronominal possessives are heads in the D projection. Adjectives are generated in designated specifier positions dominating NP, which I label αP following Julien (2002).

(21) a. a very interesting lecture on noun phrases

4According to Giusti (1993) up to seven adjectives can be combined.

11

(13)

b. DP

D a

αP

AP

very A^′

A interesting

α^′

α NP

N lecture

PP

on noun phrases In the next section I turn to the properties of noun phrases in Serbian, outlining the structure I will be assuming in the paper.

2.2 The Structure of Serbian Noun Phrases

2.2.1 Prenominal elements

Material that can precede the noun in Serbian includes the following elements:

quantifiers (22-a), demonstratives (22-b), possessives (22-c) and attributive adjectives (22-d). All prenominal elements agree with the head noun in case, number, and gender.

(22) a. sve all-N.F.Pl

knjige books-N.F.Pl

‘all books’

b. ova

this-N.F.Sg knjiga book-N.F.Sg

‘this book’

c. Jovanova Jovan’s-N.F.Sg

knjiga book-N.F.Sg

‘Jovan’s book’

d. stara old-N.F.Sg

knjiga book-N.F.Sg

‘old book’

The neutral word order is illustrated in (23).

(23) a. quantifier - demonstrative - possessive - adjective

(14)

b. sve all-N.F.Pl

ove

these-N.F.Pl

Jovanove Jovan’s-N.F.Pl

stare old-N.F.Pl

knjige books-N.F.Pl

‘all of these old books of Jovan’s’

While universal quantifiers svi ‘all’, and svako ‘each/every’ typically occupy the first position in the prenominal complex and with respect to agreement features behave like other prenominal elements, other quantificational expresions such as nekoliko ‘several’, mnogo ‘many/much’, and numerals pet ‘five’ and higher have less strict distribution and impose the genitive plural form on the noun and the material preceding it.

(24) a. mnogo many-G.Pl

starih old-G.Pl

knjiga books-G.Pl

‘many old books’

b. ovih these-G.Pl

nekoliko several-G.PL

slika

pictures-G.Pl

‘these several pictures’

In the appropriate context, the order of prenominal elements can be somewhat flexible, however the following restrictions are observed. Namely, demonstratives cannot switch order with either possessives (25-a) or adjectives (25-b).

(25) a. *Petrovi Peter’s

oni those

gosti guests

‘Peter’s those guests’

b. *dosadni boring

oni those

gosti guests

‘boring those guests’

The indefinite determiners jedan/neki ‘one/some’ pattern with demonstratives.

(26) *njegov his

jedan/neki one/some

ˇclanak article

‘his a/some article’

On the other hand, possessives and adjectives can be inverted when the focus is on the adjective.⁵

5Permutation can have a semantic effect though. The example (i-a) refers to the house John formarly owned, while (i-b) refers to an object that Jovan now owns and that was formarly a house

13

(15)

(27) dosadni boring

Petrovi Peter’s

gosti friends

‘Peter’s boring friends’

Quantifiers can also appear in several positions with different effects on interpre- tation. They can either precede or follow the demonstrative. If they precede the demonstrative, the reading we get is partitive. The contrast is illustrated below.

(28) a. Prodao sold

je aux

nekoliko several

ovih these

knjiga.

books

‘He sold several of these books’

b. Prodao Sold

je aux

ovih these

nekoliko several

knjiga books

‘He sold these several books.’

Adjectives immediatelly precede the noun they modify. Furthermore they are ordered with respect to each other. We have already seen that there are cross- linguistic regularities in the relative ordering of adjectives, although there are con- siderable differences in labels used for particular classes of adjectives. The partial ordering suggested by Sproat and Shih (1991), and adopted by Cinque (1994) is given in (29):

(29) evaluating (quality) - size - colour a. English

beautiful big red ball b. German

sh¨oner gr¨osser roter Ball

Moreover, Sproat and Shih observe that in languages where all APs follow the noun, the relative order is the mirror image of that in (29), as illustrated by In- donesian below.

(from Boˇskovi´c 2002).

(i) a. bivˇsa former

Jovanova Jovan’s

ku´ca house b. Jovanova bivˇsa ku´ca

(16)

(30) bola ball

merah red

besar big

tjantik beautiful

The same ordering restrictions are observable in Serbian. Namely, adjectives denoting size must precede the ones denoting colour, whereas both classes follow the evaluating adjectives. All of these are followed by referential adjectives, such as srpska in (31).⁶Immediatelly preceding the noun are denominal adjectives, such as svadbena ‘wedding-adj’.⁷

(31) ogromna huge

bela white

srpska Serbian

svadbena wedding

torta cake Permuting the adjectives yields a deviant output.

(32) a. *svadbena wedding-adj

bela white

torta cake b. *srpska

serbian

ogromna huge

torta cake

It can thus be concluded that the ordering patterns of prenominal elements in Serbian noun phrases are more restricted than is often assumed. In the next section I turn to the status of DP in Slavic.

2.2.2 The status of DP in Slavic

The headedness of noun phrases in Slavic has been a matter of much debate in the literature. In this chapter, I have already outlined some of the arguments that led researchers to the conclusion that noun phrases are headed by a functional projection hosting determiners in languages such as English. However, in literature on Slavic it has been often argued that the existence of a DP projection on top of an NP is a matter of parametric variation, and in particular that Slavic languages (excluding Bulgarian and Macedonian) lack the DP layer (Corver, 1990;

Zlatić, 1997; Stjepanović, 1998; Boˇsković, 2002). In this section, I review some of the arguments supporting such a proposal, and conclude that the evidence from Slavic languages is not sufficient to force us into abandoning the universality of the DP-hypothesis.

The first argument concerns the observation that Serbian does not have overt

6Referential adjectives are typically the ones expressing nationality. They seem to be related to the argument of the verb and exhibit subject-like behaviour. Cinque (1994) labels them as

‘thematic’ adjectives. I use the term referential from Giorgi and Longobardi (1991).

7Zlati´c (1997) labels these adjectives as classifying. I use the term denominal, from Boˇskovi´c (2002).

15

(17)

articles, which are typically seen as instantiations of D⁰. ⁸ However, it is worth noting that the overtness of articles is in itself not a sufficient argument for the claim that Serbian noun phrases lack the DP projection. Not only has the existence of many nonovert categories been proposed and argued for in the literature, but even limiting ourselves to English, the language for which the DP-hypothesis was originally proposed, we have to assume the existence of null D in cases such as (33) below, as plurals and mass nouns do not require the presence of an overt article.

(33) a. I don’t like scary movies.

b. She drinks coffee with milk.

The second argument concerns the observation that all prenominal elements in Serbian, including demonstratives and indefinite determiners corresponding to English some and one, as well as possessives, are indistinguishable from adjec- tives and should be analysed as such. The first piece of evidence refers to the agreement phenomena. Namely, determiners and possessives pattern with adjectives with respect to agreement, that is they agree in number, gender and case with the head noun.⁹ A partial case paradigm is given below:

(34) a. nekog

some-G/Acc-M-Sg

dosadnog

boring-G/Acc-M-Sg

ˇcoveka man b. nekom

some-Dat/Loc-M-Sg

dosadnom

boring-Dat/Loc-M-Sg

ˇcoveku man c. nekim

some-Instr-M-Sg

dosadnim

boring-Instr-M-Sg

ˇcovekom man

Following Corver (1990, 1992), Boˇskovi´c (2002) further supports this claim by arguing that, just like adjectives, the elements in question can occur in a predica- tive position of a copula construction (35-a), are able to cooccur (35-b), and have relatively free order (35-c).

(35) a. Ova

this

knjiga book

je is

moja.

*my/mine b. ta

that moja my

slika picture

8Although I am using Serbian as the representative language, the arguments presented here extend to other Slavic languages as well, apart from Bulgarian and Macedonian which are the only Slavic languages with overt articles.

9Agreement markers on adjectives do not always coincide with those on nouns. This is further discussed later in this chapter.

(18)

c. (i) Petrovi Peter’s

dosadni boring

gosti guests (ii) dosadni

boring

Petrovi Peter’s

gosti guests

Although, the presented evidence indicates that Serbian determiners and possessives show certain adjectival properties, it does not necessarily lead to the conclusion that Serbian NPs lack a DP projection. Moreover, on closer investigation the behaviour of prenominal elements seems to indicate that they should not be treated uniformly and that they, in fact occupy different positions in the extended projection of NPs.

First of all, we have already seen that the order of prenominal elements is not as free as it seems at first. Namely, determiners cannot switch order with either possessives or adjectives. I will take this to mean that the determiners in Serbian occupy the specifier position of DP (or Dem(onstrative)P as in the analysis of Scandinavian DPs in Julien (2002)).¹⁰ As far as possessives are concerned, I will assume that they appear in Spec of Poss(essive)P.¹¹ All other orders, I am assuming, are derived by movement.

Furthermore, the fact that possessives and determiners cooccur can be captured by assuming that determiners, possessives and adjectives occupy designated specifier positions of different functional projections. The conclusion is also supported by the fact that in many unrelated languages determiners and possessives do cooccur, such as in Norwegian, Hungarian, Italian, Modern Greek etc.

(36) a. a

the te you

kalap-od

hat (Hungarian) b. la

the mia my

penna

pen (Italian)

The strongest argument for assuming the lack of DP in article-less Slavic languages probably comes from extraction phenomena. Serbian allows extraction of prenominal elements out of NPs in violation of Ross’s (1967) Left Branch Condi- tion. Thus, (37-a) is grammatical in Serbian, whereas the corresponding English translation is ill-formed, and the whole DP has to be moved to sentence initial position in English, as illustrated in (37-b).

(37) a. Cijegˇ whose

je aux.

on he

gosta guest

istukao?

beaten

*‘Whose has he beaten up guest?’

10Whether determiner-like elements in Serbian are base-generated in SpecDP, or end up there as a result of movement, possibly through AgrP, is irrelevant for my purposes.

11I return to the issue of possessives in the next section.

17

(19)

b. Whose guest has he beaten up?

The contrast between English and Serbian with respect to the LBC violations has mostly been attributed to the presence vs absence of determiners (Corver 1990, 1992, Boˇskovi´c 2002). I return to the discussion of left-branch extractions in greater detail in the following chapters, where I argue that the possibility of extraction does not correlate with the presence of articles, and moreover that the constructions in question do not in fact involve the extraction of a prenominal element at all, thus obviating the need to eliminate the DP projection in Slavic NPs.

Considering that the DP layer has become established as a functional projection dominating at least some NPs in all languages, let us investigate the advan- tages of assuming that it is projected in Slavic NPs as well.

First of all, adopting the DP-hypothesis for Slavic allows us to retain the paral- lelism between the verbal and nominal domain, in that both VPs and NPs are seen as dominated by functional layers. Such an approach is strongly supported by cross-linguistic evidence. This in turn implies that the presence of DP is not subject to parametric variation, rather the projection of DP is considered a universal property, independent of the presence of a lexical item occupying the head of the projection. That is a welcome outcome. As we have seen, following Longobardi’s (1994) influential work, it is standardly assumed that only DPs can appear in argument positions, that is as subjects, objects and complements of a prepositions. On the other hand, bare NPs are limited to non-argument positions, such as predica- tives, vocatives and exclamatives. The difference in their behaviour is argued to reside in the presence of D which carries a referential index. ¹² If the main function of D is to provide a referential index, then this property of D should not be subject to parametric variation. Borer (2003) argues that if that were the case, and we allow for some languages to compute the reference in a completely different way, this would attribute a radically different computational systems to different grammars, and weaken the existence of UG as a foundation for all computational systems.

A particularly strong evidence for adopting the DP-hypothesis in Serbian comes from the noun/pronoun asymmetries. Namely, Progovac (1998) observes that in Serbian certain intensifying adjectives precede nouns, but must follow pronouns as illustrated below.

12See Baker (2003), and Pereltsvaig (2004) for arguments that NPs lacking a DP layer can function as arguments. This however does not imply that a DP projection is absent. In particular, Pereltsvaig argues that the presence of DP is necessary in Russian to account for a range of empirical data.

(20)

(38) a. (i) I and

samu alone

Mariju Mary

to that

nervira.

irritates

‘That irritates even Mary.’

(ii) *I and

Mariju Mary

samu alone

to that

nervira.

irritates b. (i) I

and

nju/mene her/me

samu alone

to that

nervira.

irritates

‘That irritates even her/me.’

(ii) *I and

samu alone

nju/mene her/me

to that

nervira.

irritates

Similar contrasts are noted by Longobardi (1994) for Italian, where interestingly, the pronoun can precede the adjective only if the article is missing, which suggests that D position is the landing site.

(39) a. *La the

sola only

lei she

si refl

e has

presentata.

showed

‘Only she showed up.’

b. Lei sola si e presentata.

c. *Sola lei si e presentata.

Progovac (1998) takes this to indicate that pronouns in Serbian move to D in overt syntax, whereas nouns stay in their base-generated positions. Moreover, she argues that Serbian pronouns cannot be generated in D, but rather surface there as a result of movement. The conclusion is based on the observation that pronouns show overt morphology not present on nouns. It was already shown that adjectives agree with nouns in gender, number and case. Sometimes these agreement markers on adjectives and nouns are not identical, and in that case adjectives show heavier agreement. Interestingly, pronouns surface bearing this heavier adjectival agreement. Consider as an illustration, the dative forms of nouns and pronouns.¹³ (40) a. (i) lep-om(u)

handsome-Dat.M.Sg

ˇcoveku man-Dat.Sg

13The agreement markers on noun and adjectives in plural typically coincide, except in the genitive form:

(i) a. lep-ih pretty-Gen.Pl

ljudi

people-Gen.Pl b. njih/ih

they-Gen

19

(21)

(ii) njemu/mu he-Dat b. (i) lep-oj

pretty-Dat.F.Sg ˇzeni

woman-Dat.Sg (ii) njoj/joj

she-Dat c. (i) lep-im(a)

pretty.Dat.Pl

ljudima people-Dat.Pl (ii) njima/im

they-Dat

Progovac argues that morphological properties of pronouns and adjectives support the existence of another functional projection below D, which she labels AgrP.

Pronouns overtly move through this projection to D, whereas nouns procrastinate their movement until LF, and thus do not surface with the same agreement pattern.

Considering that a DP layer is taken to be universal in the present framework, and since there are no convincing arguments against its existence, I will assume that the DP-hypothesis is valid even for languages lacking overt articles, such as Serbian. However, contrary to English where I concluded that determiners are heads, I will assume that in Serbian determiner-like elements occur in the specifier position of DP.

2.2.3 Prenominal Possessives

Following Abney’s work, subsequent studies have postulated an even more artic- ulated structure within noun phrases, including additional functional layers between D and N. As was already discussed, extending the functional structure is inevitable once we adopt Cinque’s (1994) analysis of adjective placement. Evi- dence for existence of a different kind of functional head is put forward by Sz- abolcsi (1994). On the basis of the Hungarian data, she argues that there is a Possessor phrase present within DP. As illustrated below, possessors in Hungarian can surface between the determiner and the noun, triggering agreement in person and number features, which is expressed in the form of a suffix on the possessed NP.

(41) a. a

the te you

kalap-ja-i-d hat-poss-pl-2sg

‘your hats’

b. a the

Mary Mary-Nom

kalap-ja-i-0 hat-poss-pl-3sg

‘Mary’s hats’

(22)

Possessors in Hungarian appear below the determiner suggesting that they are located below D in the extended projection. The fact that the possessor surfaces bearing nominative case indicates that it appears in a structural subject position, corresponding to the position of subjects in the finite clause. On this view, D is regarded as parallel to C in the verbal domain.

I will assume that possessors in Serbian should receive a similar analysis, in particular that they appear in the specifier position of PossP, projected below the DP. Prenominal possessives in Serbian are formed from unmodified, singular, an- imate nouns. As was already discussed, in the literature on Serbian they were treated as adjectives, along with other prenominal elements. However, treating possessives on a par with adjectives leaves some aspects of their behaviour unexplained. On the one hand, possessives pattern with adjectives with respect to agreement, i.e. they agree with the head noun in case, gender and number. How- ever, with respect to binding possibilities possessives show nominal properties.

Let me illustrate the latter point first, before turning to adjectival properties of possessives.

Serbian uses the following pronouns to express reflexivity: the reflexive pro- noun sebe, the clitic reflexive pronoun se, and the possessive reflexive svoj. Ser- bian reflexives are subject-oriented, in that they favour clausal subjects as their antecedents.

(42) Petar_i Peter

je aux

dao gave

Mariji_j Mary

svoju_i/_∗_j self’s

knjigu.

book

‘Peter gave Mary his book.’

Nevertheless, Zlati´c (1997) observes that possessives are able to bind anaphors as illustrated in the following example:

(43) Petar_i Peter

je aux

sluˇsao listened

Marijino_j Mary’s

opisivanje description

svoje_∗i/j self’s

majke.

mother

‘Peter listened to Mary’s description of her mother.’

The noun opisivanje in (43) belongs to the class of complex nominals in the sense of Grimshaw (1990). It has been argued in the literature that complex nominals embed verbal functional projections, with possessives acting as subjects (see Alexiadou, 2001; Schoorlemmer, 1998). This means that at some level of derivation possessives within Serbian nominals function as subjects, which enables them to bind the reflexive. Note that with non-complex nominals the reflexive has to be

21

(23)

bound by the clausal subject.

(44) Peter_i Peter

je aux.

proˇcitao read

Marijinu_j Mary’s

knjigu book

o about

sebi_i/∗j. self

‘Peter read Mary’s book about him.’

If possessives are treated uniformly as adjectives, their behaviour with respect to binding is left unexplained. Notice that even referential adjectives cannot act as syntactic subjects, as shown in (45-b) where a referential adjective is not able to bind a reflexive in the object position.

(45) a. John’s destruction of himself

b. ??the American destruction of themselves

In light of these facts, I will assume that possessives are generated as subjects within complex nominals, and subsequently move, possibly through an agreement projection, to Spec of PossP. The possessor phrase is projected on top of adjectives, but below the projection in which demonstratives appear,namely DP, thus reflecting the neutral word order.¹⁴ SpecPossP is therefore seen as a structural licensing position, equivalent to SpecIP in clauses. The possessive marking, which in Serbian surfaces as -ov/-in suffix on the possessive noun, can be seen as a morphological reflex of this formal licensing. Schoorlemmer (1998) argues that possessors show properties of structurally licensed elements; they can have different thematic roles, but only one possessor can appear within a DP. For instance, in complex nominals possessives can be interpreted as either an agent, or a theme. In example (46), where the possessive is the only argument of the noun, it is actually ambiguous between the two interpretations:

(46) Jovanovo Jovan’s

sasluˇsavanje interrogation

je aux.

trajalo lasted

satima.

hours

‘The interrogation of/by Jovan lasted for hours.’

However, if both the agent and the theme argument are present, the theme role must be expressed as a postnominal genitive NP. The agent and theme cannot both appear as possessors (47-b).

14Notice also that the behaviour of possessives provides evidence for assuming that word- formation processes are done in syntax, rather than in a separate morphological component.

(24)

(47) a. Jovanovo Jovan’s

osumnjiˇcenih suspects

je aux

trajalo lasted

satima.

hours

‘Jovan’s interrogation of the suspects lasted for hours.’

b. *Moje my

Jovanovo Jovan’s

je aux

trajalo lasted

satima hours

intended: ‘My interrogation of Jovan lasted for hours.’

The ungrammaticality of (47-b) cannot be due to the semantic restriction that

‘possession’ can only be expressed once, since neither of these arguments actually expresses semantic possession. The ill-formedness of (47-b) is also unexpected if possessives are treated as adjectives, the latter typically being iterable.

We have seen that in contrast to English, possessives in Slavic occur in a predicate position of a copular construction, which was used as an argument in favour of treating Slavic possessives as adjectival. The relevant example is repeated below.

(48) Ova this

knjiga book

je is

moja.

*my/mine

On the disjoint analysis of possessives and adjectives assumed here, the grammaticality of (48) cannot be due to the fact that possessives are adjectives. Notice that English possessives can occur predicatively, but in such contexts a longer form mine must be used. This can be taken to indicate that English possessives show weak/strong opposition as suggested by Cardinaletti (1998). In fact, she argues that Italian prenominal possessives, which have traditionally been treated as adjectives, are licensed in the prenominal ‘subject’ position, corresponding to Spec,AgrS in the clausal domain. Alternatively, Schoorlemmer (1998) argues that the difference between English-type languages, and Serbian-type languages resides in the ability of Poss to carry a definiteness feature. She observes that only languages that do not allow articles with possessors and indefinite possessor DPs have a special form in ellipsis. According to Schoorlemmer these three properties can be explained under the assumption that there is a PossP projection equivalent to IP in clauses, and the head of this phrase has a variable feature [def], which forces it to move to D. If an article is inserted in D, movement of Poss-to-D cannot occur and the derivation crashes due to the unchecked strong [def] feature.

The obligatory raising to D thus accounts for the lack of articles in possessive constructions in languages such are English, German, Dutch, and French, as well as the fact that examples with prenominal possessors are necessarily definite. In the absence of [def], Poss movement does not take place. As a consequence, possessors freely occur with articles and other determiner-like elements, and con-

23

(25)

structions with prenominal possessors can in fact be indefinite, as in the following examples from Bulgarian and Italian.

(49) a. Edin a/one

negov his

prijatel friend b. Un

a/one suo his

amico friend

Schoorlemmer assumes that the morphological shape of the possessor is sensitive to whether the possessor is sister to Poss or just the trace of Poss. In elliptical constructions, Poss is pro, which raises to Poss, and having the feature [def], is able to check the feature of Poss, precluding it from moving all the way to D. In types of languages where the form of possessor is always the same, the sister of the possessor is always Poss, since due to to the lack of [def], Poss never moves all the way to D.

I thus conclude that in order to account for the different behaviour of English and Serbian possessors, we are not committed to treating Serbian possessive forms as adjectives. In fact, some aspects of their behaviour remain unexplained if we adopt such a proposal. I will thus maintain that possessors should be treated as appearing in PossP, at least at some point in the derivation, while the differences in their behaviour should be sought elsewhere.

2.2.4 The placement of adjectives

The discussion so far has led me to the conclusion that prenominal elements cannot be treated uniformly, in other words they should not all be treated as adjectives occupying multiple specifier positions of NP.¹⁵

I have already argued that attributive adjectives in English are generated in specifier positions of functional projections dominating NP, thus essentially adopting the proposal of Cinque (1994). Considering all the arguments in support of this conclusion, I will assume the same analysis for attributive adjectives in Ser- bian. Thus, I take them to be generated in what I have labelledαP, dominating the NP.

(50) a. veliki big-M.Sg

agresivni

aggressive-M.Sg galeb

seagull-M.Sg

‘big aggressive seagull’

15Adjectives are placed within the NP in the analysis of Boˇskovi´c (2002), Corver (1990, 1992), Zlati´c (1997).

(26)

b. αP

veliki α^′

α αP

agresivni α^′ α NP

galeb

The proposed structure enables us to formally state the ordering restrictions on attributive adjectives, assuming that there is universal ordering of functional projections in whose specifiers adjectives are generated. Moreover, on the approach to extraction phenomena argued for in the following chapters, the fact that adjectives are not adjoined to NP, but rather surface in distinct functional projections, will become crucial. The observation that adjectives, along with other prenominal elements, show agreement with the head noun does not imply that they all need to be generated within the NP projection. The agreement can be established by either moving the elements in question through an agreement projection, or more plausibly the agreement does not require overt movement at all, and it can be established between the noun and prenominal elements in their base positions, which I take to be functional positions on top of NP (cf. Chomsky 2000, 2001).

Similar assumptions are made by Julien (2002), who argues that agreement in Scandinavian DP is established without movement.

Bringing all the observations together, I am assuming the following structure of noun phrases in Serbian.

(51) a. ovaj this

njegov his

brbljivi talkative

sused neighbour

‘this talkative neighbour of his’

25

(27)

b. DP

ovaj D^′

D PossP

njegov Poss^′

Poss αP

brbljivi α^′ α NP

sused

To conclude, I will basically be assuming that the structure of noun phrases is the same in English and Serbian, and in particular that Serbian NPs, just like English contain a DP functional layer. The goal of the following chapters will be to justify this assumption by dispensing with the main argument against DP status of Serbian NPs, namely the difference in extraction possibilities between the two languages. I will show that both in English and in Serbian prenominal constituents are not extracted from within the DP, while in both languages it is possible to remove DP-complements under certain conditions. Before giving the analysis of Sebian split phrases that does not rely on the presence vs absence of D, I turn first to the overview of extraction possibilities.

(28)

3 Extraction Phenomena

In this section, I present the relevant empirical data regarding extraction possibilities from DPs in Serbian, which will be the main focus of this thesis. I first explore different extraction patterns of prenominal elements, and then briefly comment on the availability of extracting postnominal constituents.

3.1 Prenominal elements

It was noted by Ross (1967) that movement of prenominal elements from within the noun phrase is ungrammatical in many languages. In order to account for inaccessibility of these constituents, Ross (1967) formulated the so called Left Branch Condition (LBC), which blocks extraction of determiners, possessors, and adjectival phrases in languages such English and Dutch, illustrated below (Dutch examples are from Corver 1990).

(52) English

a. *Which did you like film?

b. *Whose did you crash car?

c. *Lazy he hates people.

d. *How much did he drink beer?

(53) Dutch a. *Welk

which zag saw

jij you

hond van Jan?

dog of John

‘Which dog of John’s did you see?’

b. *Jans John’s

heb have

ik I

boek book

gezien!

seen

‘John’s book I have seen.’

c. *Hoe interessante how interesting

heeft has

Jan John

een lezing a lecture

gegeven?

given

If a left branch constituent is the target of movement, the entire noun phrase must be carried along in order to get a grammatical output. In other words, LBC forces obligatory pied-piping of the dominating noun phrase, in the manner illustrated below.

(54) a. Which film did you like?

b. Whose car did you crash?

(29)

Extraction Phenomena

c. How much beer did he drink?

However, already Ross (1967) observed that nonextractability of left branch constituents is not a universal property. Certain languages freely violate the Left Branch Condition. It is a well-known fact that prenominal elements are accessi- ble to extraction in Slavic languages, such as Russian, Polish, Czech and Serbian.

Thus, the sentences corresponding to the English and Dutch ill-formed examples are completely grammatical in Slavic.¹⁶ Consider the following Czech and Polish examples, taken from Corver (1990).

(55) Czech a. Jak

how

siln´eho strong

videl saw

Jan John

muˇze man

‘How strong a man did John see?’

b. Jakou which

ˇcte reads

Petr Peter

knihu?

book

‘Which book does Peter read?’

(56) Polish a. Kt´ore

which Jan John

namalowal painted

obrazy?

paintings

‘Which paintings did John paint?’

b. Czyje whose

on he

po˙zyczyl borrowed

kalosze?

galoshes

‘Whose galoshes did he borrow?’

Serbian patterns with Polish and Czech in this respect. Any prenominal element can be moved out of the noun phrase. The possibility of extracting a quantifier (57-a), demonstrative (57-b), possessive (57-c), and adjective (57-d) is exemplified below.

(57) a. Koliko how.many

su aux

otpustili fired

radnika?

workers

16Boˇskovi´c (2002) points out that Bulgarian and Macedonian, the only Slavic languages with overt articles, disallow LBE. He takes this to strongly support Corver’s assumptions. I return to this issue in chapter 6, where I show that Corver’s claim cannot be maintained.

(30)

b. Koju which

su aux

zatvorili closed

fabriku?

factory

c. Cijuˇ whose

su aux.

objavili published

knjigu?

book

‘Whose book did they publish?’

d. Kakav what.kind

su aux

kupili bought

stan?

flat

‘What kind of flat did they buy?’

Alternatively, the fronted left branch constituent can pied pipe the whole noun phrase, as in English and Dutch.

(58) a. Koliko how.many

radnika workers

su aux

otpustili?

fired b. Koju

which

fabriku factory

su aux.

zatvorili?

closed c. Cijuˇ

whose

knjigu book

su aux

objavili?

published d. Kakav

what.kind stan flat

su aux

kupili?

bought

The examples so far involved the extraction of interrogative phrases from within the object position. The following sentences illustrate that DP-splits do not arise only in wh-movement contexts.

(59) a. Andri´cevu Andri´c’s

su aux

preveli translated

knjigu.

book

‘They translated Andri´c’s book.’

b. Stare old

su aux

sruˇsili tore.down

ku´ce houses

‘They tore down old houses.’

Left branch extractions are not necessarily confined to a single clause. The target of the movement can be a left branch constituent in the embedded clause. In the following examples, the interrogative phrases koji ‘which’, and koliko ‘how much’, are fronted across a clausal boundary.

29

(31)

(60) a. Koji which

si aux

ˇcuo heard

da that

je aux

auto car

slupao?

crashed

‘Which car did you hear he crashed?’

b. Koliko how.much

misliˇs think

da that

je aux

piva beer

popio?

drunk

‘How much beer do you think he drank?

Subject DPs also belong to the set of nodes which are transparent for subextrac- tion. This is illustrated in (61-a), where the pronominal dative clitic nam, and the auxiliary clitic je intervene between the demonstrative and the noun. The second example shows that the noun phrase need not be separated.

(61) a. Ovaj this

nam us-cl

je aux-cl

predsednik president

obe´cao promised

ve´ce higher

plate.

salaries

‘This president promised us higher salaries.’

b. Ovaj this

predsednik president

nam us-cl

je aux-cl

obe´cao promised

ve´ce higher

plate.

salaries

Zlati´c (1997) argues that sentences with clitics intervening cannot be used as a reliable evidence that left-branch elements are actually removed from the subject noun phrase. However, this is only the case if we adopt an analysis of clitic placement which employs PF reordering of elements, such as Halpern’s (1995) account.

On such a view, noun phrases are split up in the PF component. The clitic is sentence initial in the output of syntax. Prosodic Inversion (PI) then places the clitic after the first accented word. The ability of clitics to break up phrasal constituents was thus considered the strongest argument for PF reordering. However, insofar as the analysis to be proposed here is on the right track, it will provide further support to syntactic accounts of clitic placement by showing that split-XPs are the result of syntactic movement.¹⁷

17Notice that while the grammatical (i-a) cannot be derived by PI, (i-b) is wrongly predicted to be well-formed:

(i) a. Cijiˇ whose

je aux-cl

Marko Marko

auto car

slupao?

crashed

‘Whose car did Marko crash?’

b. *Prema towards

je aux-cl

ku´ci house

Jovan Jovan

trˇcao.

ran

For additional arguments against Halpern (1995) see Progovac (1996) and Boˇskovi´c (2001).

(32)

It is also possible to split the subject of the embedded clause. Consider the following examples:

(62) a. Koji which

misliˇs think

da that

´ce will

kandidat candidate

pobediti win

na at

izborima.

elections

‘Which candidate do you think will win the elections.’

b. Koliko how.many

misliˇs think

da that

je aux

stiglo arrived

pisama?

letters

‘How many letters do you think have arrived?’

c. Cijeˇ whose

misliˇs think

da that

su aux

knjige books

objavljene?

published

‘Whose books do you think were published?’

Starke (2001) observes that while both preverbal and postverbal position are avail- able for subjects, extracting out of the preverbal position makes the sentence un- acceptable in Czech/Slovak. Extracting out of a postverbal subject is equivalent to extracting out of an object. The contrast is illustrated below.

(63) a. Kolik how.many

myslis think

ze that

prislo arrived

dopisu?

letters b. *Kolik

how.may

myslis think

ze that

dopisu letters

prislo?

arrived

The same is not true of Serbian, that is preverbal subjects are not opaque for extraction. In fact, all cases of extraction in Serbian actually improve if the remnant of the subextraction precedes the verb.¹⁸

(64) a. Koliko how.many

misliˇs think

da that

je aux

pisama letters

stiglo?

arrived b. Koju

which su aux

oni they

fabriku factory

zatvorili?

closed

c. Cijuˇ whose

je aux

on he

knjigu book

preveo?

translated

‘Whose book did he translate?’

18According to my Russian informants, Russian patterns with Serbian in this respect.

31

(33)

The following examples show that extraction of left branch elements from within NP complements is disallowed.

(65) a. (i) On

he je aux

pozajmio borrowed

knjigu book

ovog this

studenta.

student

‘He borrowed a book of this student.’

(ii) *Kojeg which

je aux

on he

pozajmio borrowed

[knjigu book

studenta]?

student

‘Of which did he borrow a book student.’

b. (i) On he

je aux

izgubio lost

adresu address

njihove their

sestre.

sister

‘He lost the address of their sister.’

(ii) * ˇCije whose

je aux

on he

izgubio lost

[adresu address

sestre]?

sister

‘Of whose did he lose address sister?’

In example (65-a-ii), the wh-word kojeg ‘which’ corresponding to the demon- strative in the declarative sentence, cannot be moved out of the postnominal NP complement. The same is true of the possessive interrogative in (65-b-ii). NP complements thus form opaque domains for extraction.

All the examples presented so far involved the extraction of a left branch constituent from within nominal phrases. I now turn to the conditions under which prepositional phrases can be split.

PP Splits It is a well-known fact that Slavic languages strictly prohibit preposition stranding. The complement of the preposition cannot be fronted leaving the preposition behind as in (66-b), nor can the preposition itself be moved to the sentence initial position, stranding the complement as in (66-c). Thus, even though Slavic languages are known for great freedom in ordering of constituents, no permutation of elements that separates the preposition from its complement is permitted (the example is taken from Abels (2003)).

(66) a. Jovan Jovan

je aux

trˇcao ran

prema towards

velikoj big

ku´ci.

house

‘Jovan ran towards a big house.’

b. *Velikojku´ci je Jovantrˇcaoprema.

(34)

c. *Prema towards

je aux

Jovan Jovan

trˇcao ran

velikoj big

ku´ci.

house

Nevertheless, PPs can be discontinuous under certain conditions. In the following examples, a left branch constituent is fronted together with the preposition.

Alternatively, the whole prepositional phrase can be pied-piped, as illustrated in (67-c).

(67) a. Prema towards

velikoj big

je aux

Jovan Jovan

ku´ci house

trˇcao.

ran b. Sa

with kojim which

je aux

Nataˇsa Nataˇsa

profesorom professor

razgovarala?

talked

‘Which professor did Nataˇsa talk to?’

c. Sa with

kojim which

profesorom professor

je aux

Nataˇsa Nataˇsa

razgovarala?

talked

On the other hand, the movement cannot affect the preposition and the noun, leaving the prenominal modifiers behind.

(68) *Prema ku´ci je Jovan trˇcao velikoj.

towards house aux Jovan ran big

Finally, consider the following examples where the prepositional phrase con- tains a left branch modifier. In that case, the modifier of the preposition must be carried along with the fronted part of the PP.

(69) a. On

he je aux

uˇsao entered

pravo straight

u into

veliku big

sobu.

room

‘He went straight into the big room.’

b. Pravo straight

u into

veliku big

je aux

on he

uˇsao entered

sobu.

room c. *U

into veliku big

je aux

on he

uˇsao entered

pravo straight

sobu.

room

The discussion so far addressed only the conditions under which prenominal constituents can be extracted from either nominal or prepositional phrases. Before turning to the analysis of the presented facts, I will briefly explore the mobility of postnominal elements.

33

(35)

3.2 Postnominal elements

Facts concerning the extraction of postnominal elements seem to be much less clear and constant, with the amount of degradation varying considerably across speakers. In this section, I will attempt to identify some general tendencies.

Starke (2001) argues that in languages with case morphology the possibility of extraction correlates with the nature of case assigned to a DP. Specifically, only noun phrases with structural case are transparent for extraction, while inherent case creates a strong island. Consider the Czech paradigm taken from Starke (2001).

(70) a. Ktereho which_Gen

doktora doctor_Gen

to it

byla was

chyba?

fault_{N om} b. ?Ktereho

which_Gen herce actor_Gen

by would

sis you

rad gladly

koupil buy

obrazek?

picture_Acc c. *Ktereho

bys would

sis you

rad gladly

vynadal scold

priteli?

friend_Dat d. *Ktereho

se you

bojis fear

pritele?

friend_Gen

The observation seems to hold for Serbian as well, that is DPs bearing inherent case are entirely opaque for extraction. The following examples illustrate that removing an NP-complement from dative (71-a), instrumental (71-b), and genitive (71-c) DPs leads to a strong degradation.

(71) a. *[Kojih which

partija]_Gen parties

je aux

on he

verovao trust

voąama_Dat? leaders

‘Leaders of which parties did he trust?’

b. *[Kojih which

studenata]_Gen students

se aux

ponosio proud.of

radovima_Instr papers

‘Papers by which students was he proud of?’

c. *[Kojih which

ljudi]_Gen people

se aux

plaˇsio be.afraid

optuˇzbi_Gen? accusations

‘Accusations by which people was he afraid of?’

However, judgements concerning the grammaticality of extraction from DPs with structural case seem to vary considerably. Zlati´c (1994) argues that neither NP nor wh-extraction is possible from NP complements of nouns in Serbian, giving the following examples:

Nominal subextractions and the structure of NPs in Serbian and English